Qualcomm AI Engine Direct - SqueezeNet Enablement (#3748)

Summary:
Summary
- OSS model enablement: squeezenet

Pull Request resolved: #3748

Reviewed By: kirklandsign

Differential Revision: D57896519

Pulled By: cccclai

fbshipit-source-id: 2afacd4272891f0914a39855e3e7d67d166d2076

Author: haowhsu-quic

backends/qualcomm/tests/test_qnn_delegate.py

@@ -1524,6 +1524,7 @@ def test_dino_v2(self):
         ]
         if self.host:
             cmds.extend(["--host", self.host])
+
         p = subprocess.Popen(cmds, stdout=subprocess.DEVNULL)
         with Listener((self.ip, self.port)) as listener:
             conn = listener.accept()
@@ -1566,6 +1567,39 @@ def test_esrgan(self):
             self.assertGreaterEqual(msg["PSNR"], 24)
             self.assertGreaterEqual(msg["SSIM"], 0.8)
 
+    def test_squeezenet(self):
+        if not self.required_envs([self.image_dataset]):
+            self.skipTest("missing required envs")
+
+        cmds = [
+            "python",
+            f"{self.executorch_root}/examples/qualcomm/oss_scripts/squeezenet.py",
+            "--dataset",
+            self.image_dataset,
+            "--artifact",
+            self.artifact_dir,
+            "--build_folder",
+            self.build_folder,
+            "--device",
+            self.device,
+            "--model",
+            self.model,
+            "--ip",
+            self.ip,
+            "--port",
+            str(self.port),
+        ]
+        if self.host:
+            cmds.extend(["--host", self.host])
+
+        p = subprocess.Popen(cmds, stdout=subprocess.DEVNULL)
+        with Listener((self.ip, self.port)) as listener:
+            conn = listener.accept()
+            p.communicate()
+            msg = json.loads(conn.recv())
+            self.assertGreaterEqual(msg["top_1"], 40)
+            self.assertGreaterEqual(msg["top_5"], 70)
+
 
 class TestExampleScript(TestQNN):
     def required_envs(self, conditions=None) -> bool:

examples/qualcomm/oss_scripts/squeezenet.py

@@ -0,0 +1,160 @@
+# Copyright (c) Qualcomm Innovation Center, Inc.
+# All rights reserved
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import json
+import os
+import sys
+from multiprocessing.connection import Client
+
+import numpy as np
+
+import torch
+from executorch.backends.qualcomm.quantizer.quantizer import QuantDtype
+from executorch.examples.qualcomm.scripts.utils import (
+    build_executorch_binary,
+    make_output_dir,
+    parse_skip_delegation_node,
+    setup_common_args_and_variables,
+    SimpleADB,
+    topk_accuracy,
+)
+
+
+def get_dataset(dataset_path, data_size):
+    from torchvision import datasets, transforms
+
+    def get_data_loader():
+        preprocess = transforms.Compose(
+            [
+                transforms.Resize(256),
+                transforms.CenterCrop(224),
+                transforms.ToTensor(),
+                transforms.Normalize(
+                    mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+                ),
+            ]
+        )
+        imagenet_data = datasets.ImageFolder(dataset_path, transform=preprocess)
+        return torch.utils.data.DataLoader(
+            imagenet_data,
+            shuffle=True,
+        )
+
+    # prepare input data
+    inputs, targets, input_list = [], [], ""
+    data_loader = get_data_loader()
+    for index, data in enumerate(data_loader):
+        if index >= data_size:
+            break
+        feature, target = data
+        inputs.append((feature,))
+        targets.append(target)
+        input_list += f"input_{index}_0.raw\n"
+
+    return inputs, targets, input_list
+
+
+if __name__ == "__main__":
+    parser = setup_common_args_and_variables()
+
+    parser.add_argument(
+        "-d",
+        "--dataset",
+        help=(
+            "path to the validation folder of ImageNet dataset. "
+            "e.g. --dataset imagenet-mini/val "
+            "for https://www.kaggle.com/datasets/ifigotin/imagenetmini-1000)"
+        ),
+        type=str,
+        required=True,
+    )
+
+    parser.add_argument(
+        "-a",
+        "--artifact",
+        help="path for storing generated artifacts by this example. "
+        "Default ./squeezenet",
+        default="./squeezenet",
+        type=str,
+    )
+
+    args = parser.parse_args()
+
+    skip_node_id_set, skip_node_op_set = parse_skip_delegation_node(args)
+
+    # ensure the working directory exist.
+    os.makedirs(args.artifact, exist_ok=True)
+
+    if not args.compile_only and args.device is None:
+        raise RuntimeError(
+            "device serial is required if not compile only. "
+            "Please specify a device serial by -s/--device argument."
+        )
+
+    data_num = 100
+    inputs, targets, input_list = get_dataset(
+        dataset_path=f"{args.dataset}",
+        data_size=data_num,
+    )
+    pte_filename = "squeezenet_qnn"
+    instance = torch.hub.load(
+        "pytorch/vision:v0.10.0", "squeezenet1_1", pretrained=True
+    )
+    build_executorch_binary(
+        instance.eval(),
+        (torch.randn(1, 3, 224, 224),),
+        args.model,
+        f"{args.artifact}/{pte_filename}",
+        inputs,
+        skip_node_id_set=skip_node_id_set,
+        skip_node_op_set=skip_node_op_set,
+        quant_dtype=QuantDtype.use_16a16w,
+    )
+
+    if args.compile_only:
+        sys.exit(0)
+
+    # setup required paths accordingly
+    # qnn_sdk       : QNN SDK path setup in environment variable
+    # artifact_path : path where artifacts were built
+    # pte_path      : path where executorch binary was stored
+    # device_id     : serial number of android device
+    # workspace     : folder for storing artifacts on android device
+    adb = SimpleADB(
+        qnn_sdk=os.getenv("QNN_SDK_ROOT"),
+        artifact_path=f"{args.build_folder}",
+        pte_path=f"{args.artifact}/{pte_filename}.pte",
+        workspace=f"/data/local/tmp/executorch/{pte_filename}",
+        device_id=args.device,
+        host_id=args.host,
+        soc_model=args.model,
+    )
+    adb.push(inputs=inputs, input_list=input_list)
+    adb.execute()
+
+    # collect output data
+    output_data_folder = f"{args.artifact}/outputs"
+    make_output_dir(output_data_folder)
+
+    adb.pull(output_path=args.artifact)
+
+    # top-k analysis
+    predictions = []
+    for i in range(data_num):
+        predictions.append(
+            np.fromfile(
+                os.path.join(output_data_folder, f"output_{i}_0.raw"), dtype=np.float32
+            )
+        )
+
+    k_val = [1, 5]
+    topk = [topk_accuracy(predictions, targets, k).item() for k in k_val]
+    if args.ip and args.port != -1:
+        with Client((args.ip, args.port)) as conn:
+            conn.send(json.dumps({f"top_{k}": topk[i] for i, k in enumerate(k_val)}))
+    else:
+        for i, k in enumerate(k_val):
+            print(f"top_{k}->{topk[i]}%")